A research note · Rev 1.0 · Apr 2026

How the InsiderScore works.

A 7-factor framework, designed by a CFA charterholder, applied to every SEC Form 4 and STOCK Act disclosure.
Greg Taylor, CFA·Transparent scoring·Audited nightly

Academic researchhas documented insider-buying alpha for four decades — from Seyhun's 1986 Journal of Financial Economics paper to Cohen, Malloy, and Pomorski's more recent “decoding insider information” work. The signal exists. The problem is noise: the SEC processes tens of thousands of Form 4 filings a year, and most are routine — option grants, 10b5-1 scheduled sales, tax-driven dispositions.

InsiderBrief's job is to separate the informative from the noisy. We do this with a transparent 7-factor scoring model for corporate insiders, and a parallel 6-factor model for members of Congress and their spouses. Every filing we ingest is scored on a 0–100 scale within 15 minutes of hitting EDGAR.

The score is not a prediction. It is a measure of how informative a filing is, relative to the 40-year academic literature.

Below: the full rubric. No black boxes, no machine-learning confounders. Every factor is computable from the filing itself plus a handful of market variables (price, 52-week range, SPY reference). Weights were calibrated against historical Form 4 data and will be re-tuned as the live track record matures.

Corporate model

Seven factors · 100 points

SEC Form 4
1
Transaction type
Purchases signal conviction; sales are noisy.
Open-market purchases (Form 4 Code P) score the full 20. Sales (Code S) score 0 here — they're tracked, but the other six factors carry the signal weight. Academic research consistently shows insider buying beats the market; selling does not.
Weight20
2
Insider seniority
CEOs and directors have information asymmetry.
Executive and director-level filers score higher than 10%-beneficial-owner filings or indirect holdings. A CEO buying their own stock is the highest-signal role we observe; entity-level 10% owners carry weight, but less than direct C-suite action.
Weight15
3
Dollar size
Bigger purchases indicate stronger conviction.
Bucketed: >$5M = 15, >$1M = 12, >$500K = 9, >$100K = 6, >$50K = 3, else 1. Filters applied upstream (purchases below $25K and sales below $1M are excluded), so every scored trade clears a minimum materiality bar.
Weight15
4
Relative size
Size vs. the insider's existing position.
A $1M purchase by an insider who already owns $100M of stock is a 1% position add — weaker than the same dollar purchase by someone whose total holding triples. Scored against shares-owned-following from the Form 4 table.
Weight15
5
Cluster signal
Multiple insiders, same ticker, 7-day window.
Applied after initial scoring. Detects when two or more distinct insiders at the same company transact in the same direction within 7 days. Cluster buys are the single most predictive pattern in academic insider-trading literature.
Weight15
6
Timing context
Post-earnings and post-drawdown trades score higher.
Purchases immediately after earnings (when insiders have full information and open windows) score more than mid-quarter trades. Purchases during broad-market drawdowns (SPY down 5%+ weekly) earn extra points — insiders buying into fear is a classic contrarian tell.
Weight10
7
Contrarian signal
Buying into weakness; selling into strength.
Purchases when the stock is 20%+ off its 52-week high score more — insiders are betting against the broader pessimism. Sales into 52-week-high territory are equally informative on the bearish side.
Weight10
Congressional model

Six factors · 100 points

STOCK Act · PTR
1
Owner type
Self > Joint > Spouse > Dependent.
Legislators themselves signal more than spouse or dependent trades. Self = 15, Joint = 12, Spouse = 6, Dependent = 3.
Weight15
2
Dollar size
Midpoint of the disclosure range, bucketed.
STOCK Act disclosures are reported in bands ($1,001–$15,000, $15,001–$50,000, etc.). We use the midpoint for scoring: larger bands score higher, capped at 25 points.
Weight25
3
Disclosure delay
Faster disclosure = more credible.
The STOCK Act requires filing within 45 days. Early filers (under 10 days) score the full 15; late filers (40+ days) earn 3. Pattern of habitually-late filing from the same legislator is flagged separately.
Weight15
4
Committee signal
Relevant committee = information access.
A member of Armed Services buying a defense prime, or Energy & Commerce trading a pharma name, scores higher than a random sector trade. We maintain a mapping of committees to the GICS sectors they oversee.
Weight20
5
Transaction direction
Purchases beat sales; exchanges zero out.
Parallel to the corporate model: purchase = full 15, sale = 0, partial purchase = 10. Option exercises and in-kind exchanges score 0 — they don't carry directional conviction.
Weight15
6
Historical accuracy
Track record of individual legislators.
We score each legislator on a rolling 24-month forward-return track record. Consistently accurate filers earn a bonus; consistently-wrong filers lose points. Recalibrated quarterly.
Weight10
Output

From raw score to tier

Once all factors are summed, the raw 0–100 score is bucketed into three public tiers. Tier boundaries were chosen to align with forward-return inflection points observed in historical data; the live track record on /performance updates these nightly.

75+
STRONG
Top ~10% of scored filings.
Large purchases from senior insiders, often in clusters, often into drawdowns. These are the filings we lead the morning brief with.
TierStrong
60–74
NOTABLE
Meaningful, but not headline.
Single-insider purchases with good context (post-earnings, post-drawdown), or committee-relevant congressional disclosures. Read these in the daily digest.
TierNotable
40–59
ROUTINE
Clears the bar; lower conviction.
Still clears the academic bar historically, but by a narrower margin. Surfaces in lookups and performance tracking; rarely leads the brief.
TierRoutine
The goal is not to predict price. It is to tell you which of yesterday's filings are worth ten minutes of your time.

The forward-return track record for every scored signal lives on /performance. No survivorship, no rebase — if we scored it, it's there.

/to search